Peter McCrory

Yeah.

So the way that I would think about this is the 1.8 percentage point

number that we talked about relies on this thing called holton's theorem which is a sort of standard way to add up efficiency gains without making too many assumptions about how those efficiency gains emerge in general it's being able to do a particular task faster and given the structure of the economy today what would that mean overall if you add up those those efficiency gains

We ran the survey of Claude users recently, where we asked them about their fears and hopes with AI.

And one of the things that emerged in that survey was a lot of discussion around the labor market impact and also the productivity gains that people are experiencing.

More so than being able to do things faster, people were more often likely to indicate that Claude helps them do more.

So broadening of scope.

that broadening of scope looks more like labor augmentation than labor automation.

And in that sense,

I think that that might be an important aspect of the productivity gains that we ultimately might experience.

I guess, you know, this goes back to this point that the technology is broadly applicable, and we also want to be good stewards of the technology.

Or at least I think that that's like part of the goal.

The thing that motivates me in our research is we're not just putting data and research out there because it's interesting.

We're doing it because the impact on society and on the economy will be shaped by the technology.

by the decisions that we make as much by the technological capabilities themselves.

And so in that sense, it's wise to have conversations with all stakeholders

who might be affected by the technology so that we can steward our role responsibly.

I think the way that I think about mythos is that it is, you know, the thing that make Claude very good at helping me very quickly do sophisticated statistical analysis and econometric analysis for my purposes or help me build interactive dashboards quite quickly.

That's the same capability that allows the model to

identify vulnerabilities in critical infrastructure and software systems and as we talked about before these models are doubling roughly every four to seven months in terms of one notion of their capabilities and so in that sense it's just consistent with this exponential that we've been on